Atmos. Meas. Tech., 13, 6459–6472, 2020https://doi.org/10.5194/amt-13-6459-2020© Author(s) 2020. This work is distributed underthe Creative Commons Attribution 4.0 License.

Improved cloud detection over sea ice and snow during Arcticsummer using MERIS dataLarysa Istomina1,2, Henrik Marks2, Marcus Huntemann1,2, Georg Heygster2, and Gunnar Spreen2

1Alfred-Wegener-Institut, Helmholz Zentrum für Polar und Meeresforschung, 27570 Bremerhaven, Germany2Institute of Environmental Physics, University of Bremen, 28357 Bremen, Germany

Correspondence: Larysa Istomina (larysa.istomina@awi.de)

Received: 30 October 2019 – Discussion started: 16 January 2020Revised: 3 September 2020 – Accepted: 10 September 2020 – Published: 2 December 2020

Abstract. The historic MERIS (Medium Resolution ImagingSpectrometer) sensor on board Envisat (Environmental Satel-lite, operation 2002–2012) provides valuable remote sensingdata for the retrievals of summer sea ice in the Arctic. MERISdata together with the data of recently launched successorOLCI (Ocean and Land Colour Instrument) on board Sen-tinel 3A and 3B (2016 onwards) can be used to assess thelong-term change of the Arctic summer sea ice. An importantprerequisite to a high-quality remote sensing dataset is an ac-curate separation of cloudy and clear pixels to ensure lowestcloud contamination of the resulting product. The presenceof 15 visible and near-infrared spectral channels of MERISallows high-quality retrievals of sea ice albedo and melt pondfraction, but it makes cloud screening a challenge as snow,sea ice and clouds have similar optical features in the avail-able spectral range of 412.5–900 nm.

In this paper, we present a new cloud screening methodMECOSI (MERIS Cloud Screening Over Sea Ice) for the re-trievals of spectral albedo and melt pond fraction (MPF) fromMERIS. The method utilizes all 15 MERIS channels, includ-ing the oxygen A absorption band. For the latter, a smileeffect correction has been developed to ensure high-qualityscreening throughout the whole swath. A total of 3 years ofreference cloud mask from AATSR (Advanced Along-TrackScanning Radiometer) (Istomina et al., 2010) have been usedto train the Bayesian cloud screening for the available limitedMERIS spectral range. Whiteness and brightness criteria aswell as normalized difference thresholds have been used aswell.

The comparison of the developed cloud mask to the oper-ational AATSR and MODIS (Moderate Resolution ImagingSpectroradiometer) cloud masks shows a considerable im-

provement in the detection of clouds over snow and sea ice,with about 10 % false clear detections during May–July andless than 5 % false clear detections in the rest of the melt-ing season. This seasonal behavior is expected as the sea icesurface is generally brighter and more challenging for clouddetection in the beginning of the melting season.

The effect of the improved cloud screening on the MPF–albedo datasets is demonstrated on both temporal and spatialscales. In the absence of cloud contamination, the time se-quence of MPFs displays a greater range of values through-out the whole summer. The daily maps of the MPF now showspatially uniform values without cloud artifacts, which wereclearly visible in the previous version of the dataset.

The developed cloud screening routine can be applied toaddress cloud contamination in remote sensing data over seaice.

The resulting cloud mask for the MERIS operating time, aswell as the improved MPF–albedo datasets for the Arctic re-gion, is available at https://www.seaice.uni-bremen.de/start/(Istomina et al., 2017).

1 Introduction

No other surface type of satellite imagery has the unique fea-tures of bright, reflecting white snow surface. The task ofsnow detection therefore would be an easy task in the ab-sence of clouds. However, the snow spectral signature (e.g.,Warren, 1982) is also a feature of water and especially ofice clouds (Kokhanovsky, 2006). Possible snow impurities,snow grain size differences and liquid water content createfine differences between many snow types (Warren, 1982),

Published by Copernicus Publications on behalf of the European Geosciences Union.

6460 L. Istomina et al.: Improved cloud detection over sea ice and snow during Arctic summer

but in general the spectra of snow and cloud are similar inthe visible and near infrared, with the difference occurringbeyond 1 µm (e.g., channels at 1.6, 3.7, 11 and 12 µm).

For MERIS data with a spectral range from 412.5 to900 nm, cloud detection over snow and sea ice is a challeng-ing task. However, the advantage of MERIS – its 15 spectralbands within this relatively small spectral range – makes itespecially suitable for the melt pond fraction (MPF) retrievalover the Arctic sea ice, which needs a quality cloud screeningroutine.

Although most of the field campaigns and in situ mea-surements of the sea-ice-covered Arctic ocean are performedduring Arctic summer (e.g., an overview in Istomina et al.,2015), the links and feedbacks between the rapidly evolvingsea ice surface, the atmosphere and the under-ice ecosystemare multifold (Curry et al., 1996) and not yet fully under-stood. The appearance of melt ponds on sea ice during meltonset causes a drastic change of its albedo and transmittance(Nicolaus et al., 2012) which affects the surface energy bal-ance and facilitates lateral, top, bottom and internal sea icemelt, i.e., affects the sea ice volume. Only recently the sug-gestion that melt ponds during melt onset might be connectedto the sea ice area during the sea ice minimum has been pub-lished (Schröder et al., 2014). In order to understand theseprocesses, a long-term global coverage record of sea ice pa-rameters, among others also MPF, needs to be available tothe community. That is, the presented cloud screening rou-tine and the resulting MPF dataset can be used in studies ofsea ice processes and feedbacks.

To the knowledge of the authors, at the time of writing noclimate model includes melt ponds on top of sea ice. Oneof the reasons is that melt ponds, although observed in situduring many campaigns, still present a challenge for climatemodeling due to unknown global spatial distribution. Al-though reanalysis air temperature at the surface is also avail-able over sea-ice-covered Arctic ocean (e.g., Kalnay et al.,1996), MPF is not linearly linked to the air temperature butalso depends on the ice topography and its internal macro-physical properties such as density, porosity, etc. SatelliteMPF datasets of possibly global coverage are the only wayto understand not only local events but also global spatial dy-namics, which may eventually lead to successful inclusion ofmelt ponds into climate models.

Besides cloud screening for the MPF retrieval usingMERIS data, a robust cloud detection from MERIS in theArctic region may be important for (1) synergy with the othersensors on board Envisat and (2) might be applicable to sen-sors similar to MERIS, e.g., OLCI.

The cloud screening for OLCI, which is a successor ofMERIS without thermal infrared bands, presents challengessimilar to those of MERIS. OLCI data are important asa continuation of MERIS in order to provide long-termdata records of, for example, MPF. Nevertheless, the cloudscreening presented here has been developed specifically forMERIS and thus addresses the issue of cloud screening over

snow for ENVISAT sensors, e.g., SCIAMACHY (SCanningImaging Absorption spectroMeter for Atmospheric CHartog-raphY; see e.g., Schlundt et al., 2011). Of course, the ap-proach presented here can be applied to OLCI data as well.

Depending on the retrieved parameter and sensor, the ef-fect of a compromised cloud screening may be moderate (re-trievals of albedo and snow grain size within SGSP, snowgrain size and pollution amount retrieval; Wiebe et al., 2013)to drastic (aerosol retrieval, Istomina et al., 2011; MPF re-trieval, Zege et al., 2015). As the melting sea ice displays avariety of spectral behaviors in the entire range from whiteice to dark melt ponds (e.g., Istomina et al., 2013), a versa-tile forward model and retrieval which can account for sucha variability at a global spatial scale are needed. Such a re-trieval (melt pond detector, MPD) has been developed byZege et al. (2015). The MPD is a pixelwise retrieval and onlyutilizes the spectral information without additional morpho-logical or statistical criteria. As clouds do not spectrally dif-fer from most of the surfaces available during Arctic summer,so that, for example, warm water clouds may appear similarto white ice throughout most of the available spectral range(same for cirrus and fresh fine snow), the MPD can thereforemisinterpret the cloud contamination as sea ice melt. The re-sulting MPF and albedo datasets are thus strongly affected bythe residual cloud contamination. The objective of this workis to resolve this issue by means of a reliable cloud discrim-ination over snow for MERIS and to provide the datasets ofMPF, albedo and cloud mask of a better quality than currentlyavailable.

2 Sensor-specific cloud screening in remote sensing

Some sensors are better suited for the task of cloud screeningbut are not suitable for the MPD retrieval due to other limi-tations. For example, the MODIS cloud mask (Ackermannet al., 1998; Liu et al., 2004) is one of the most compre-hensive classification algorithms; however, as the MODISsensor experiences saturation in some of the visible bands(Madhavan et al., 2012), it is impossible to use these datafor the given sea ice albedo and MPF retrieval (Zege et al.,2015). As the MERIS sensor on board Envisat does not havethese limitations, it has been chosen for the retrievals of MPFand albedo. However, the choice of methodology to performcloud screening over snow and ice with MERIS is not a triv-ial task.

Three basic cloud-screening approaches applicable to aspectroradiometer data can be distinguished among the avail-able algorithms.

– Analyze time sequences of data, under the assumptionthat the short-term changes of the scene can only be in-troduced by clouds (e.g., Key and Barry, 1989; Diner etal., 1999; Lyapustin et al., 2008; Lyapustin and Wang,2009; Gafurov and Bárdossy, 2009). Such an approachassumes surfaces with a constant and pronounced struc-

Atmos. Meas. Tech., 13, 6459–6472, 2020 https://doi.org/10.5194/amt-13-6459-2020

L. Istomina et al.: Improved cloud detection over sea ice and snow during Arctic summer 6461

ture (Lyapustin et al., 2008; Lyapustin and Wang, 2009).Although the approach proved to be effective for variousnatural and artificial surfaces, it is not applicable withinthis work due to the fast-evolving nature of melt pondsand the sea ice.

– Apply a reflectance or brightness temperature absolutethreshold or their combination, e.g., ratio of reflectancesin the form of normalized difference vegetation index(NDVI). In this case, only a few channels are used (e.g.,Minnis et al., 2001; Bréon and Colzy, 1999; Lotz etal., 2009; Allen et al., 1990; Spangenberg et al., 2001;Trepte et al., 2001). The optical properties of snow inthe visible spectral range show weak spectral depen-dency. In the near infrared and infrared, however, thesnow spectrum shows the typical “snow signature”, i.e.,values decreasing due to water absorption in the nearinfrared, which also causes the dependence on the snowgrain size due to different path length and absorptionin the grains of different size. These features aid thesnow–cloud discrimination. Therefore, it is a commonpractice to use infrared channels in addition to the visi-ble for such retrievals (Spangenberg et al., 2001). In thecurrent task, the limited spectral range of MERIS doesnot allow effective usage of this approach.

– Analyze image processing and spatial variability (e.g.,Martins et al., 2002). In the case of white clouds overwhite surface, the spatial variability would mainly comefrom the difference in grain/particle size, surface rough-ness, different water phase (ice surface vs. water cloud,melting surface vs. ice cloud) and cloud shadows. Giventhe great natural variability of these parameters in bothArctic clouds and surface and the similarity of their op-tical properties in the given spectral range, this approachis prone to false detections.

Combinations of the above methods together with additionalthresholds and additional meteorological/reanalysis data arealso available. For example, the MODIS cloud detectionscheme (Ackerman et al., 1998; Liu et al., 2004) is one ofthe most comprehensive among the available cloud detec-tion schemes and is based on such a combination. This al-gorithm uses 19 out of 36 MODIS channels along with addi-tional inputs, e.g., topography and illumination observationgeometry for each 1 km pixel, land–water mask, ecosystemmaps and daily operational snow–ice products (taken fromthe NOAA, National Oceanic and Atmospheric Administra-tion and NSIDC, National Snow and Ice Data Center). Theresulting MODIS cloud mask contains four confidence levels(confident cloudy, uncertain, probably clear, confident clear)and is available as a separate daily averaged product. Un-fortunately, due to the time lag between Envisat and Terra–Aqua, the MODIS cloud mask product cannot be used forswathwise screening for the melt pond fraction retrieval.

Most of the cloud screening approaches do not focus onthe case of the snow surface; among those who do (Allenet al., 1990; Spangenberg et al., 2001; Trepte et al., 2001;Istomina et al., 2010, 2011), even a smaller fraction uti-lizes MERIS sensor for this task (Kokhanovsky et al., 2009;Schlundt et al., 2011; Zege et al., 2015; Istomina et al., 2015;Krijger et al., 2011).

The currently available cloud masks for MERIS (Zege etal., 2015; Schlundt et al., 2011, etc.) are based on NDSI-like (normalized difference snow index) indices, e.g., MDSI(MERIS Differential Snow Index). In the absence of infraredchannels these thresholds result in a residual cloud contam-ination over snow and sea ice. However, unlike most of themoderate resolving spectroradiometers, MERIS has the so-called oxygen A band (MERIS channel 11 at 761.5 nm),which can aid greatly in cloud screening over snow and ice.

MERIS oxygen A band and the smile effect

As oxygen is well mixed in the Earth atmosphere, the ampli-tude of absorption within MERIS channel 11 reflects opticalpath length of light rays received with the sensor. Effectivepath length over clouds is shorter than that over sea ice orsnow on land; that is, light reflected from clouds experiencesless absorption when traveling through the atmosphere thanlight reflected from the surface. This allows the separation ofclouds and snow or sea ice surface according to their heightin the atmospheric column. We expect this criterion to workbest for optically thick water clouds. The sensitivity to opti-cally thin clouds is expected to be small over bright surfaceslike sea ice (Preusker and Lindstrot, 2009), and clouds witha low top height would generally also have a weaker effecton the oxygen ratio. Fortunately, as in our case the Arcticsea ice surface lies uniformly at sea level and displays norelief, there is no confusion possible between clouds and sur-face in the terms of optical path length, and the only uncer-tainty might come from the sensor-specific features, i.e., thesmile effect. As MERIS is a push-broom sensor, its chan-nels are susceptible to the smile effect usual for this type ofsensor, which occurs due to a small variation in the centralwavelength across the MERIS swath. The artifacts appear asalong-track stripes within the swath and impair application ofthresholds or retrievals. The oxygen ratio approach withoutthe smile correction has been used by Zege et al. (2015) andIstomina et al. (2015) as a threshold in addition to classicalwhiteness and brightness criteria. The threshold comprisesthe ratio of 11 (oxygen A absorption) and 10 (oxygen A ref-erence) R11/R10< 0.27, where the value 0.27 has been de-rived from the visual analysis of several dozen MERIS scenes(see Zege et al., 2015, and Eq. 17 therein). As seen in Zegeet al. (2015) and Istomina et al. (2015), strong artifact pres-ence compromises the effective application of the oxygen ra-tio threshold.

The smile effect of MERIS has been studied (Bourg etal., 2008), and ways to correct it have been shown by

https://doi.org/10.5194/amt-13-6459-2020 Atmos. Meas. Tech., 13, 6459–6472, 2020

6462 L. Istomina et al.: Improved cloud detection over sea ice and snow during Arctic summer

Gómez-Chova et al. (2007) or Jäger (2013). The approachby Jäger (2013) greatly improves the usability of the oxy-gen ratio but does not fully remove detector-to-detector dif-ferences. A reason for this might be instrument stray light,which is not fully removed in the MERIS operational pro-cessing chain (Lindstrot et al., 2010), and that was not takeninto account by Jäger (2013).

Another available smile effect, corrections also com-prise those included in the ESA (European Space Agency)toolbox for Envisat processing, i.e., open-source packagesBEAM or SNAP (Earth Observation Toolbox and Develop-ment Platform, Sentinels Application Platform, https://www.brockmann-consult.de, last access: 30 July 2020). These cor-rections work well within the transparency window of the at-mosphere over darker surfaces but are not sufficient in theoxygen A absorption band over brighter surfaces such assnow and ice.

In this work, we suggest a smile correction for MERISband 11 which allows slight inaccuracy on the absolute valueof the top-of-atmosphere (TOA) reflectances but preservesthe relative difference between the sensor pixels, which al-lows a quantitative use of the corrected oxygen A band forcloud screening (Sect. 3.3.1).

3 Methods

The cloud screening method for MERIS data developed inthis work is specifically aimed to work well over summer seaice. It is called MECOSI (MERIS Cloud screening Over SeaIce). MECOSI utilizes the collocated AATSR data in the cen-ter part of the MERIS swath, namely the infrared channels1.6, 3.7, 11 and 12 µm, for training. In this work, we use theAATSR cloud screening developed for the aerosol retrievalover snow and ice (Istomina et al., 2010).

Currently MECOSI is being applied as preprocessing forthe retrieval of melt pond fraction and spectral albedo ofsummer sea ice (MPD). The MPD retrieval takes top-of-atmosphere reflectances of MERIS at nine channels as in-put and employs a forward model of optical properties ofthe Arctic surface and an iterative procedure to retrieve spec-tral albedo and melt pond fraction of a given pixel. Severalhundred field spectra of the Arctic sea ice and melt pondshave been used to constrain the input parameters of the for-ward model and to ensure realistic range of modeled sur-faces. More details on the MPD retrieval can be found inZege et al. (2015). The presented cloud screening methodcan be used for other remote sensing applications as well,e.g. for retrievals of other surface or atmospheric parametersor as a cloud mask for coarser resolving sensors on boardsame satellite platform (e.g., SCIAMACHY on Envisat).

3.1 Data used

Input for MECOSI is MERIS Level 1B observations. MERISconsists of five cameras scanning the surface of the Earth inpush-broom mode and offers 15 spectral bands from 412.5to 900 nm. The data are collected globally with a spatial res-olution of 1040 m× 1200 m at nadir. The Level 1B prod-uct provides calibrated and georeferenced TOA radiances.These are preprocessed using the software package BEAM(http://www.brockmann-consult.de/cms/web/beam/, last ac-cess: 30 July 2020).

The preprocessing includes the following.

1. The region north of 65◦ N is cut out from each orbitusing the module subset.

2. The metadata in the L1B swaths are given in a grid withreduced resolution and need to be interpolated in orderto have the data available for each pixel. This is doneusing the BandMath module. The coordinates as well assun zenith and the view zenith angles are now interpo-lated.

3. The TOA radiances are corrected and converted to re-flectances using the module Meris.CorrectRadiometry.The correction includes an equalization step to reducedetector-to-detector differences and a scheme to reducethe smile effect in all but the absorption bands 11 and15.

A cloud mask derived from AATSR data (Istomina et al.,2010) is used as a reference mask to develop and validatethe MECOSI algorithm. The AATSR instrument has beenlaunched together with MERIS aboard Envisat, and both sen-sors observe the same scene nearly simultaneously. The spa-tial resolution of AATSR is 1 km at nadir, which is similar tothe spatial resolution of MERIS. However, as AATSR has anarrower swath of 512 km, it covers only the central half ofa MERIS swath. The AATSR cloud-screening algorithm hasbeen developed for an aerosol optical thickness retrieval andis presented by Istomina et al. (2010). It exploits knowledgeabout the spectral shape of snow in visible, near-infrared andthermal infrared bands of AATSR. As intercomparisons ofcloud-screening routines are challenging due to the time dif-ference between the overflights of different satellite sensors,the validation has been performed against in situ lidar data.The comparison of the AATSR cloud mask to the micro-pulse lidar data has proven the robustness of the method(95 % correct cloudy and clear detections with the remaining5 % of cases connected to thin clouds on a sample of ∼ 100scenes). The output is a binary mask for cloud-free snow andice.

The training dataset used in this work was prepared asfollows: all AATSR swaths from May to September 2009,2010 and 2011 have been subset, transformed into TOAand co-located to the corresponding MERIS swaths using a

Atmos. Meas. Tech., 13, 6459–6472, 2020 https://doi.org/10.5194/amt-13-6459-2020

L. Istomina et al.: Improved cloud detection over sea ice and snow during Arctic summer 6463

nearest-neighbor algorithm (radius of influence 1.5 km). Asthe AATSR and MERIS data have different spatial resolu-tions, the two datasets have been gridded to a single grid (thecoarser grid of MERIS). This might have affected the pixelsat the borders of clouds in a way that earlier fully coveredpixels now become partly covered, which the binary AATSRcloud mask cannot fully reflect. Therefore we exclude thetwo-pixel border from the study.

This AATSR dataset from May to September 2009–2011was used to estimate the cloudy and clear case probabilitiesfor the given feature vector as described in the next sections.

3.2 Bayesian cloud screening

A comprehensive introduction to the theory of Bayesiancloud screening is given by Hollstein et al. (2015). The de-scribed approach can be found in detail in Marks (2015). Inthe following, P(A,B) denotes the occurrence probabilityof A under the condition that the occurrence of B and F is avector of features derived pixelwise from satellite data. If Cdenotes cloudy conditions (C – clear conditions), the proba-bility of seeing a cloudy pixel under the occurrence of F canbe written as

P(C,F )=P (F ,C) ·P(C)

P (F ,C) ·P (C)+P(F ,C

)·P(C) . (1)

Using this equation to calculate the cloud probabilityP(C,F ), we need to estimate the probabilities P(F ,C) andP(F ,C) for each possible feature vector F ∈ RN . We ac-complish this by calculating N -dimensional frequency his-tograms, one for cloud and one for clear-sky cases as flaggedin the AATSR mask. This is done for every AATSR andMERIS swath for the time period 1 May to 30 September2009. The background probability P(C) is directly calcu-lated from the AATSR masks using data from the same year.Pixels outside the AATSR swath are not used in this analy-sis. The set of features for which the above procedure is beingperformed is described below.

3.3 Features and applied corrections

The selection of the features used to build the feature vectorF is the most important step during the development of thealgorithm and greatly affects the performance of the screen-ing. Hollstein et al. (2015) used a random search algorithm tofind a set of features Fi that performs best in global applica-tion. Here, however, the features are selected manually to finda set that performs best over snow-covered ice and darker,ponded ice. Additionally, correction algorithms were devel-oped to equalize the systematic dependencies on the cross-track pixel position.

3.3.1 Oxygen A ratio

The TOA ratio of the oxygen A band 11, which is located atthe oxygen absorption line at 761 nm, to band 10 at 754 nm,

which is the oxygen reference band, is used here to estimatethe absorption by oxygen in the atmospheric column abovethe reflecting surface:

rox =R11

R10. (2)

The ratio rox cannot be used directly in the feature vectorF because of dependencies on the illumination–observationgeometry and because of the smile effect artifacts. The op-tical path through the atmosphere depends on sun and viewzenith angles in both cloudy and clear cases and needs tobe accounted for. As these angles are provided in MERISLevel 1B swath data, the air mass factor can be calculated(e.g., Gómez-Chova et al., 2007). The smile effect artifactsin rox need additional consideration before it can be used forcloud screening.

In this work, we propose an empirical approach to equalizerox and decrease the influence of the abovementioned factorsacross the swath. We assume that over a statistically signifi-cant sample, the mean value of rox for a given set of condi-tions (e.g., for a given detector index, geometry) can be usedto correct the systematic across-track dependence for this setof conditions. We assume that rox depends on three parame-ters: the detector index Id which corresponds to the positionof the pixel in the detector array, the sun zenith angle θs andthe viewing zenith angle θv. Id gives a pixel’s position in thesensor array and allows compensation for the spectral smileeffect. The sun zenith angle θs and the viewing angle θv allowthe estimation of the optical path in the atmosphere which isdirectly dependent on the oxygen absorption. The seasonalnature of rox dependence on surface reflectance, for exam-ple, at channel 779 nm presents a challenge of statisticallynon-uniform bins of vastly different sample sizes and wasnot included in the correction scheme. The residual rox de-pendence on the surface reflectance is less than 2 % (Fig. 2)and does not prevent the application of the cloud screeningroutine. Assuming θsum = θs+θv, we obtain a set of data vec-tors:

M ={(rox,θsum,Id)i

}, i ∈ I. (3)

The set I denotes the indices of all pixels in one swath. Pixelswith the same detector index Id are selected from the set Mand corresponding subsets are built:

Mj= {(rox,θsum,Id) ∈M|Id = j} . (4)

These subsets Mj are then processed separately. The ratio isbinned as follows:

Rjθ =

{rox|(rox,θsum,Id) ∈M

j , θ ≤ θsum < θ + δ}. (5)

The bin width δ is set to 1/4◦. The sets Rjθ are calculated formany swaths K , typically all summer data of 1 year. Thenthe mean value of rox is calculated for each one of these sets:

rjθ =mean

{rox|rox ∈

K⋃k

(Rjθ

)k

}. (6)

https://doi.org/10.5194/amt-13-6459-2020 Atmos. Meas. Tech., 13, 6459–6472, 2020

6464 L. Istomina et al.: Improved cloud detection over sea ice and snow during Arctic summer

Finally, a fifth-order polynomial is fitted to the averaged val-ues for each separate detector index j to achieve smooth andcontinuous correction functions f j :

f j = fit{rjθ

}, (7)

which in addition are functions of θsum. The correction is ap-plied pixelwise by evaluating f and subtracting the resultingvalue from the oxygen A ratio. The corrected ratio is thenused as a feature in the cloud screening algorithm.

It must be noted that as the further calculation of cloudprobabilities for the given detector indices and values of roxhappens in the space of corrected rox only, the absolute am-plitude of rox is not important for our application and is notpreserved within the described approach. Instead, the relativedifference between the scattering events at the surface and atthe cloud are equalized throughout the swath and thus madeavailable for cloud screening.

The above-described approach has been performed overall MERIS swathes to above 65◦ N for the time range from1 May to 30 September 2009. This sample is considered tobe statistically significant in terms of variety of surface andcloud types and their seasonal behavior under a variety ofobservation–illumination geometries for all detector indices.

3.3.2 MERIS Differential Snow Index

The MERIS Differential Snow Index (MDSI) is defined asnormalized difference of the TOA reflectances at 865 and885 nm:

Fsi =R13−R14

R13+R14. (8)

It exploits the drop in spectral reflectance of snow and ice atthe given wavelengths to aid discrimination of snow and icefrom clouds (Schlundt et al., 2011). The systematic cross-track variation is less pronounced than that for the oxygenA ratio, and no dependence on the observational geometry isexpected; i.e., it is assumed to be the same for both spectralbands R13 and R14. Therefore, we use a simplified correctionscheme: the mean value of Fsi is calculated for each detectorindex using swaths from the summer 2009. Open-water pix-els have been removed using two thresholds on channels 12and 13 as described by Schlundt et al. (2011). As before, toremove the systematic across-track variability, the obtainedmean values are subtracted from Fsi for each detector index.

3.3.3 Brightness and whiteness

Many types of clouds have a higher reflectance than snowin the near infrared and they usually show a white spectrum.The usefulness of these two features to detect clouds has beenshown in Gómez-Chova et al. (2007) and the same definitionsare used here. The brightness b is a spectral integral over thereflectance. As the spectral resolution of the sensor is quite

coarse with only 13 used channels, the brightness can be rep-resented by the following equation:

b =1

λmax− λmin

∑iεI

ri+1+ ri

2(λi+1− λi) . (9)

Here, λ denotes the center wavelength of a MERIS band andI is the set of used bands. The absorption bands 11 and 15 areexcluded from the calculation; hence, we use bands 1–10 and12–14 to calculate the overall brightness b. The whiteness wof the spectrum is measured by the deviation of the radiancesfrom the brightness b. With ei = |ri − b|, the equation is

w =1

λmax− λmin

∑iεI

ei+1+ ei

2(λi+1− λi) . (10)

Note that small values forw correspond to a flat and thereforewhite spectrum.

3.4 Evaluation

The cloud probabilities for each given set of features(Sect. 3.2) were compiled into binary masks in order tocompare the results to the binary AATSR cloud masks.The masks are created by normalizing the cloud probabil-ity P(F,C) to the range [0,1] and splitting the dataset at aprobability threshold of 0.45 to introduce binary values. Anoperation of morphological closing and opening was thenapplied to the cloud and snow or sea ice pixels in orderto remove single pixels. The binary MECOSI and AATSRcloud masks are used to filter out clouds in the MPD swathdata. No co-location or interpolation is necessary for thisstep because both algorithms, the MECOSI cloud screeningand MPD, process identical MERIS swaths, and the AATSRcloud masks were gridded to the MERIS grid. The compar-ison of the three cloud masks, as well as illustration of sep-arate features of the feature vector F as well as their correc-tions, is given in the next section.

4 Results

4.1 Oxygen A correction

An example of the influence of the oxygen A correction de-scribed in Sect. 3.3.1 is presented in Fig. 1. The jumps at thetransition between the five detectors of MERIS, visible asvertical stripes in the uncorrected ratio (Fig. 1b), are stronglyreduced by applying the correction (Fig. 1c). The influenceof low sun elevation, which causes the dark top left cornerin the uncorrected ratio, is much less apparent. Also, thereare no pronounced artifacts introduced by the discrete look-up table (Sect. 3.3.1) used for the correction, as the correctedratio is a rather smooth image. Very bright pixels, e.g., cloudedges visible in Fig. 1a, are darker and more apparent afterapplying the correction.

Atmos. Meas. Tech., 13, 6459–6472, 2020 https://doi.org/10.5194/amt-13-6459-2020

L. Istomina et al.: Improved cloud detection over sea ice and snow during Arctic summer 6465

Figure 1. Reflectance at 779 nm (a), uncorrected oxygen A ratio (b) and corrected oxygen A ratio used as a feature in the cloud screening (c).Shown is a 2450× 1121 pixel part of Envisat orbit 37475 from 1 May 2009 with the New Siberian Islands at the bottom and parts of theCanadian Archipelago at the top. Land, open water and invalid pixels are white.

Figure 2. Along-track mean of the corrected oxygen A ratio. Foreach time period, the mean is calculated from 100 randomly se-lected swaths. The vertical lines mark the transition between thefive detectors of MERIS.

Another way to investigate the effect of the correction isto study the along-track mean of the oxygen A ratio. As ex-pected, the corrected ratio is a smooth function with valuesclose to zero, if data from the whole period May to Septem-ber are considered (Fig. 2 black line). This is different forthe data from May only, where we find small jumps betweenthe detectors (Fig. 2 red line). Moreover, there is a negativeslope in the along-track mean, which implies that pixels atthe right side of the swath tend to be darker than the ones onthe left side. For the data of July, we find a reverse sign situa-tion (Fig. 2, blue line). This seasonal dependence is expecteddue to the illumination–observation geometry change in thecourse of summer; however, these artifacts are minimal andstill allow a high-quality cloud detection using the oxygen AMERIS band.

4.2 Comparison to AATSR cloud mask

We first investigate whether the MECOSI algorithm can re-produce the AATSR cloud mask for the year 2009 used forthe algorithm training. As AATSR data also contain thermalinfrared bands, in which the snow and ice surface is virtuallya blackbody, the cloud detection with AATSR shows goodreliability in the Arctic (Istomina et al., 2010) and can beused as a reference in this study. Figures 3 and 4 show twoexamples of the MECOSI cloud probability, one for the typi-cal situation at the beginning of the melt season in May withbright, snow-covered ice (Fig. 3) and one for darker, pondedice at the peak of the melt season in July (Fig. 4). In bothcases, the cloud probability (Figs. 3b and 4b) correspondsto the AATSR mask (Figs. 3c and 4c). Most clouds visiblein the TOA reflectance images (Figs. 3a and 4a) are promi-nent with significantly higher cloud probabilities. No distinctdifference in cloud probability is visible across the swathand dependencies on the acquisition geometry or detector-specific properties appear to be well compensated. However,closer inspection reveals several cases of false negatives, likethe semitransparent clouds over landfast ice which cannot bediscriminated from clear-sky regions by their cloud probabil-ity (red arrow in Fig. 3a). The opposite case is shown with ablue arrow in Fig. 3a, where low ice concentrations close tothe coast were falsely detected as high cloud probability.

To quantify the performance of the algorithm, we studythe distribution of cloud probability for clear-sky and cloud-covered pixels in the AATSR mask (Fig. 5). For cloud-covered pixels, we find that nearly 85 % percent show acloud probability greater than the background probabilityP(C)= 0.86, and the distribution drops sharply towardssmaller cloud probabilities (Fig. 5a). Visual inspection shows

https://doi.org/10.5194/amt-13-6459-2020 Atmos. Meas. Tech., 13, 6459–6472, 2020

6466 L. Istomina et al.: Improved cloud detection over sea ice and snow during Arctic summer

Figure 3. Reflectance at 779 nm (a), cloud probability (b) and corresponding AATSR mask (c) for 14 May 2009 with Svalbard at the bottomleft corner. Land, open water and invalid pixels are white. The red arrow points to missed clouds and the blue one marks wrongly screened-outclear-sky pixels (orbit number 37666).

Figure 4. Reflectance at 779 nm (a), cloud probability (b) and corresponding AATSR mask (c) for 31 July 2009 (orbit number 38778). Theblue arrow marks a region with wrongly screened-out clear-sky pixels, although a thin cloud cover is possible.

that probabilities smaller than P(C) are almost always corre-lated to semitransparent cloud over snow-covered ice or opti-cally thin clouds. The distribution for clear-sky pixels is lessdistinct (Fig. 5b). It drops towards higher cloud probabilities,which is expected, but 6 % show a cloud probability higherthan P(C) and cannot be reliably discriminated from clouds.The majority of these 6 % are the challenging case of bright,snow-covered sea ice during the beginning of the melt seasonand fresh snow during fall freeze-up; hence such incorrectlyhigh cloud probability is rarely found for darker ice with meltponds on top. Most of these false positives are connected to

cloud-like values of the MDSI feature Fsi, which may poten-tially occur for fresh snow with fine grains. The extremelyhigh albedo of such a surface will compromise the rox fea-ture and prevent correct detection.

We compare the MECOSI binary mask to the AATSRreference mask to study the temporal behavior of the algo-rithm’s performance and to investigate the accuracy of thebinary mask. By comparing all swaths from May to Septem-ber 2009, we find that, with reference to the AATSR cloudmask, 92.51 % of the MERIS pixels are classified correctlyand the remaining 7.49 % split up into 4.64 % missed cloud

Atmos. Meas. Tech., 13, 6459–6472, 2020 https://doi.org/10.5194/amt-13-6459-2020

L. Istomina et al.: Improved cloud detection over sea ice and snow during Arctic summer 6467

Figure 5. Distribution of MECOSI cloud probability for AATSRcloud pixels (a) and AATSR clear-sky pixels (b) for May to Septem-ber 2009.

Figure 6. Time series of daily mean classification rates for 2009. Asan example, a value of 0.9 for cloud means that 90 % of the cloudpixels in the AATSR mask are correctly classified as cloud-coveredand the remaining 10 % are missed clouds.

and 2.85 % missed clear-sky pixels. The temporal behavior ofthe detection rates is presented in Fig. 6. The algorithm worksbest in July, with detections rates around 0.9 for both clear-sky and cloud pixels, and the performance is only slightlyworse in June. However, we find a considerably worse detec-tion rate for clear-sky regions in May, August and Septemberwith values close to 0.6 and below. This indicates that morethan 40 % of the pixels marked as clear sky in the AATSRmask are falsely screened out in the MECOSI binary mask.The detection rate for cloud steadily increases during Juneand July up to almost 1.0 at the end of the melt season. Thisincrease is due to the state of ice surface, which gets darkerover time and makes the detection of semitransparent cloudeasier.

The binary cloud mask derived from MECOSI cloud prob-ability is compared to the independent AATSR mask fromtwo other years. By comparing over 3.8× 109 pixels from2010 and 2011, we find that 90.50 % (90.65 % for 2011) ofthe pixels are correctly classified, which is about 2 % less

Figure 7. Along-track mean and standard deviation of cloud prob-ability for 2010. Vertical lines mark the transition between the fivedetectors of MERIS.

than for 2009. Thereby 5.85 % (5.92 %) are missed cloud and3.64 % (3.42 %) are wrongly screened-out clear-sky pixels.

4.3 Extension beyond AATSR swath and comparisonto MODIS cloud fraction

The accuracy of the MECOSI algorithm outside of the cen-ter half of the swath is difficult to assess because of the lackof appropriate reference data. Visual inspection of MERISimages from 2009 to 2011, which have been superimposedwith the binary cloud mask, gives the general impression thatthe accuracy is considerably good throughout the full swath.The several cases of semitransparent clouds in May and earlyJune 2010 are more frequently missed in the upper right quar-ter of the swath. The reason for this is somewhat small valuesin the corrected oxygen A ratio; a tendency towards smallervalues on the right side of the swath is also observable in May2009 (Fig. 2). The along-track mean of cloud probability forthe year 2010 also gives slightly smaller values at the rightside of the swath, as Fig. 7 shows, and the standard devia-tion σ increases. However, the differences across the swathare small (± 0.017 for the mean and ± 0.02 for σ ) and aremainly linked to different characteristics of the five detectorsof MERIS, as the jumps at the transitions and the linear be-havior for the center detectors show.

To further investigate the performance outside of theAATSR swath as well as the overall accuracy, we compareMECOSI binary cloud mask, gridded to a 1◦ constant an-gle grid, to MODIS cloud fraction (Ackermann et al., 2008)data from May to September 2010. Thereby, we use eitherthe full MERIS swath, center half or the outside quarters(Fig. 8a, b and c, respectively). We find a good agreementwith the MODIS data in all three cases. If the full MERISswath is used (Fig. 8a), the comparison of over 6.7× 105 gridcells gives a root-mean-square deviation RMSD= 0.18 anda difference of means D=−0.02, which indicates thatthe MECOSI algorithm tends to retrieve slightly highercloud fraction. The numbers for the central part of theswath (Fig. 8b) are very similar, with RMSD= 0.19 andD=−0.03, but the number of grid cells N = 5.0× 105 is

https://doi.org/10.5194/amt-13-6459-2020 Atmos. Meas. Tech., 13, 6459–6472, 2020

6468 L. Istomina et al.: Improved cloud detection over sea ice and snow during Arctic summer

Figure 8. Comparison of daily gridded MECOSI and MODIS cloud fraction using the full MERIS swath (a), the center half (b) or the outsidequarters (c) for the gridded MECOSI fraction. Period is May to September 2010.

smaller because of the restricted spatial coverage. For theoutside quarters, we find again almost equal parameters withRMSD= 0.19, D=−0.01 and N = 4.6× 105, although aslight pixel displacement is seen (compare top left and bot-tom right corner of Fig. 8b and c).

4.4 Influence on the melt pond fraction retrieval

Finally, we study the influence of different cloud-maskingschemes on the retrieved MPF. Figure 9 shows an exampleof using the original cloud screening built into the MPD al-gorithm (Zege et al., 2015; Istomina, 2017), as well as the ef-fect of additionally applying the MECOSI and AATSR cloudmasks. It is evident that both the MECOSI and the AATSRcloud mask (Fig. 9b and c) are much more restrictive than theMPD cloud masking scheme (Fig. 9a). The spatial coverageis significantly reduced and regions which are not screenedout correspond well to a MODIS cloud fraction below 50 %(Fig. 9d). Differences between using the MECOSI and theAATSR cloud mask are mostly due to the limited spatial cov-erage of AATSR (e.g., the larger pole hole).

A time series of the Arctic-wide mean MPF for all threecloud masking schemes is presented in Fig. 10. The spatialcoverage has been restricted to the area seen by AATSR.

For all three years 2009 to 2011, we find evident differ-ences between the original MPD product and the two im-proved products with additional cloud masking. The mostprominent one is the significantly higher (up to 0.08 increase)mean MPF in July when additional cloud screening is ap-plied. In May and September, however, the additional screen-ing results in slightly smaller mean MPF. This behavior is ex-pected because the MPD algorithm retrieves values of around0.15 MPF for opaque clouds, so that immense cloud contam-ination in the original MPD product reduces the MPF valuerange of the time series towards this wrong MPF value.

If we focus on the differences between AATSR andMECOSI cloud mask (dark red and blue in Fig. 10), we findthat both masks lead to a similar MPF time series. Using theMECOSI mask results in slightly higher MPF in May, whichis possibly caused by some omitted clouds. The main advan-

tage of the MECOSI cloud mask over AATSR is the largerspatial coverage of the latter (compare Fig. 9b and c).

5 Discussion

The results show that the MECOSI algorithm discriminatesclouds from summer sea ice with good accuracy. WithMECOSI, over 90 % of the pixels are classified correctly,when compared to the AATSR reference.

Comparison to the independent MODIS daily cloud frac-tions shows good agreement with the developed MECOSImask both in the center part of the MERIS swath whereAATSR data are available for training and on the outsideedge of the swath (Fig. 8). There is no evidence that thequality of the algorithm performance worsens towards theedges of the swath. The variation in mean cloud probabilityand its standard deviation across the swath are dominated bydetector-to-detector differences and show no change towardsthe edges of the swath (Fig. 7). Therefore, we conclude thatthe results of the comparison to the AATSR cloud mask are,in general, valid for the full MERIS swath.

The quality of the MECOSI cloud mask for both clear andcloudy cases is the best in June and July, when the rapid meltonset and first pond drainage events happen on the Arcticsea ice (Fig. 6). Bright fresh snow compromises MECOSIcloud screening and leads to some false detections in May.The oxygen A ratio is well suited for improving the detectionover fresh snow. The proposed correction scheme equalizesthe ratio reasonably well (compare Fig. 1b and c). However,the detector-to-detector artifacts indicate some residual influ-ences of the spectral smile effect, surface albedo and instru-ment stray light which were also not fully removed by theproposed correction scheme.

The cloud detection rates at the end of the melting seasonin August–September are close to 100 %. The not-as-gooddetection of clear cases might be connected to the reducednumber of such scenes at the end of the melting season, ashumidity and cloudiness increase and the ice cover decreaseswith the minimum ice extent, typically in the first weeks of

Atmos. Meas. Tech., 13, 6459–6472, 2020 https://doi.org/10.5194/amt-13-6459-2020

L. Istomina et al.: Improved cloud detection over sea ice and snow during Arctic summer 6469

Figure 9. Gridded melt pond fraction with MPD cloud mask (a), MECOSI cloud mask (b), AATSR cloud mask (c) and MODIS daytimemean cloud fraction (d), 20 June 2009.

September. For our specific application, i.e., retrieving sur-face parameters, it is important to screen out possibly allclouds as they bias the retrieval result. Wrong detections ofclear cases as cloudy are less critical as this just reduces thespatial coverage of the product but does not affect the re-trieved values.

Consequently, the MECOSI cloud screening improves thequality of the MPD MPF and albedo product. By reduc-ing the amount of cloud contamination, we find consistentlyhigher pond fraction in the period from mid-June to mid-August for all 3 years (Fig. 10). The cloud-contaminated pix-els are no longer used as input into the MPD retrieval, andthe resulting MPF dataset contains unbiased MPF and albedovalues. The thus improved resulting dataset can be used forfurther applications, such as assimilation into or validation ofclimate and melt pond models.

6 Summary

In this work, we present MECOSI, a new cloud-screeningroutine for MERIS specifically developed for use over Arc-tic summer sea ice. Comparison to the independent MODIScloud mask shows that the available summer Arctic MPF

and spectral sea ice albedo product from MERIS (Zegeet al., 2015; Istomina et al., 2015) are significantly cloud-contaminated (compare Fig. 9a and d). The cloud-screeningmethod presented here has been developed to improve thequality of the MPF and albedo datasets.

The developed cloud-masking routine utilizes all 15MERIS channels and a reference AATSR cloud mask to cal-culate probabilities of cloudy and clear cases for a given setof features:

– oxygen A absorption and reference ratio (additionallycorrected for smile effect),

– MERIS normalized difference snow index,

– brightness and whiteness criteria.

The dependencies on the illumination–observation geometryand the position of the pixel in the array of detectors, i.e., thedetector index, have been accounted for as well. To calculatethe cloudy and clear probabilities, a dataset of every AATSRand MERIS swath from 1 May to 30 September 2009 havebeen used to ensure a representative sample of the sea ice,snow and cloud conditions.

https://doi.org/10.5194/amt-13-6459-2020 Atmos. Meas. Tech., 13, 6459–6472, 2020

6470 L. Istomina et al.: Improved cloud detection over sea ice and snow during Arctic summer

Figure 10. Influence of different cloud mask on Arctic-wide mean melt pond fraction for 2009–2011. The means are calculated from griddedmelt pond fraction data and coverage is restricted to the area seen by AATSR. Days with less than 100 grid cells to compare or missingAATSR data are excluded.

The developed cloud mask shows a considerable improve-ment over the old MPD cloud mask. The quality of cloud de-tection of the new algorithm is close to the reference AATSRcloud mask, whereas MERIS does not have the infraredchannels which aid in the snow–cloud discrimination. TheMECOSI cloud detection quality also remains high near theedges of the MERIS swath where no AATSR training datawere available. Comparison to the reference AATSR and in-dependent MODIS cloud masks shows that the application ofMECOSI has greatly increased the quality of the MPD prod-ucts on both spatial (Fig. 9) and temporal (Fig. 10) scales.

The advantage of MECOSI over, for example, MODISdaily cloud fraction product is that it enables accurate cloudscreening of swath MERIS data over snow and sea ice, whichwas not possible with the old version of the cloud screeningused in the MPD retrieval. The developed cloud-screeningroutine can be applied to remote sensing data of sea ice sur-faces in both cold and melting conditions.

The developed cloud mask for MERIS over the summerArctic sea ice, as well as the improved datasets of the meltpond fraction and spectral albedo for the entire MERIS op-eration time are available at the server of the Universityof Bremen (https://www.seaice.uni-bremen.de/start/; Istom-ina et al., 2017).

Data availability. The developed MECOSI cloud mask for MERISover the summer Arctic sea ice, as well as the improved datasets

of the melt pond fraction and spectral albedo are available at https://www.seaice.uni-bremen.de/start/ (Istomina et al., 2017). The orig-inal MPD MPFs are available at https://seaice.uni-bremen.de/data/meris/gridded/ (Istomina, 2017).

Author contributions. LI conceived of the original idea and pro-vided its first implementation, worked with the MERIS and AATSRdata to prepare the training and validation of the algorithm, providedthe MERIS MPF datasets, analyzed and interpreted the results, andoutlined and wrote the manuscript with contributions from all au-thors. HM advanced the method and defined the optimal set ofBayesian features as well as performed the training and validationof the algorithm with MODIS data. MH contributed to the con-cept and implementation of the smile correction and the Bayesianmethod and contributed to the discussions. GH and GS supervisedthe project and aided the development of the algorithm and the dis-cussion of the results. All authors provided critical feedback on themanuscript.

Competing interests. The authors declare that they have no conflictof interest.

Acknowledgements. The authors express gratitude to ESA andNASA for providing MERIS, AATSR and MODIS operational andhigher-level products and to Brockmann Consult for providing theopen-source software packages BEAM and SNAP.

Atmos. Meas. Tech., 13, 6459–6472, 2020 https://doi.org/10.5194/amt-13-6459-2020

L. Istomina et al.: Improved cloud detection over sea ice and snow during Arctic summer 6471

The authors are grateful to the anonymous reviewers and the ed-itor for their effort and valuable comments on the manuscript.

Financial support. This research has been supported by theEU project SPICES (grant no. 640161) and by the DeutscheForschungsgemeinschaft (SPP 1158 project REASSESS, grantno. 424326801).

The article processing charges for this open-access publica-tion were covered by the University of Bremen.

Review statement. This paper was edited by Alexan-der Kokhanovsky and reviewed by two anonymous referees.

References

Ackerman, S. A., Strabala, K. I., Menzel, W. P., Frey, R. A., Moeller,C. C., and Gumley, L. E.: Discriminating clear sky from cloudswith MODIS, J. Geophys. Res., 103, 32141–32157, 1998.

Ackerman, S. A., Holz, R., Frey, R., Eloranta, E., Mad-dux, B., and McGill, M.: Cloud Detection with MODIS.Part II: Validation, J. Atmos. Ocean. Tech., 25, 1073–1086,https://doi.org/10.1175/2007JTECHA1053.1, 2008.

Allen, R. C., Durkee, P. A., and Wash, C. H.: Snow/Cloud Discrim-ination with Multispectral Satellite Measurements, J. Appl. Me-teorol., 29, 994–1004, 1990.

Bourg, L., D’Alba, L., and Colagrande, P.: MERIS Smile EffectCharacterization and Correction, Tech. rep., European SpaceAgency, available at: https://earth.esa.int/documents/700255/707220/MERIS_Smile_Effect.pdf (last access: 30 July 2020),2008.

Bréon, F.-M. and Colzy, S.: Cloud detection from the spacebornePOLDER instrument and validation against surface synoptic ob-servations, J. Appl. Meteorol., 38, 777–785, 1999.

Curry, J. A., Rossow, W. B., Randall, D., and Schramm,J. L.: Overview of Arctic cloud and radiation characteris-tics, J. Climate, 9, 1731–1764, https://doi.org/10.1175/1520-0442(1996)009<1731:OOACAR>2.0.CO;2, 1996.

Diner, D., Clothiaux, E., and Di Girolamo, L.: MISR Multi-angleimaging spectro-radiometer algorithm theoretical basis. Level 1Cloud detection, Jet Propulsion Laboratory, California Instituteof Technology, Pasadena, USA, JPL D-13397, 1999.

Gafurov, A. and Bárdossy, A.: Cloud removal methodology fromMODIS snow cover product, Hydrol. Earth Syst. Sci., 13, 1361–1373, https://doi.org/10.5194/hess-13-1361-2009, 2009.

Gómez-Chova, L., Camps-Valls, G., Calpe-Maravilla, J., Guanter,L., and Moreno, J.: Cloud-Screening Algorithm for EN-VISAT/MERIS Multispectral Images, IEEE T. Geosci. Remote,45, 4105–4118, https://doi.org/10.1109/TGRS.2007.905312,2007.

Hollstein, A., Fischer, J., Carbajal Henken, C., and Preusker,R.: Bayesian cloud detection for MERIS, AATSR, andtheir combination, Atmos. Meas. Tech., 8, 1757–1771,https://doi.org/10.5194/amt-8-1757-2015, 2015.

Istomina, L.: The MPD MPF dataset, available at: https://seaice.uni-bremen.de/data/meris/gridded/ (last access: 30 July 2020),2017.

Istomina, L. G., von Hoyningen-Huene, W., Kokhanovsky, A. A.,and Burrows, J. P.: The detection of cloud-free snow-covered ar-eas using AATSR measurements, Atmos. Meas. Tech., 3, 1005–1017, https://doi.org/10.5194/amt-3-1005-2010, 2010.

Istomina, L. G., von Hoyningen-Huene, W., Kokhanovsky, A. A.,Schultz, E., and Burrows, J. P.: Remote sensing of aerosols oversnow using infrared AATSR observations, Atmos. Meas. Tech.,4, 1133–1145, https://doi.org/10.5194/amt-4-1133-2011, 2011.

Istomina, L., Nicolaus, M., Perovich, D. K.: Spectralalbedo of sea ice and melt ponds measured during PO-LARSTERN cruise ARK-XXVII/3 (IceArc) in 2012, In-stitut für Umweltphysik, Universität Bremen, PANGAEA,https://doi.org/10.1594/PANGAEA.815111, 2013.

Istomina, L., Heygster, G., Huntemann, M., Schwarz, P., Birn-baum, G., Scharien, R., Polashenski, C., Perovich, D., Zege,E., Malinka, A., Prikhach, A., and Katsev, I.: Melt pond frac-tion and spectral sea ice albedo retrieval from MERIS data– Part 1: Validation against in situ, aerial, and ship cruisedata, The Cryosphere, 9, 1551–1566, https://doi.org/10.5194/tc-9-1551-2015, 2015.

Istomina, L., Marks, H., and Huntemann, M.: MECOSI cloudscreening dataset, University of Bremen, available at: https://www.seaice.uni-bremen.de/start/ (last access: 30 July 2020),2017.

Jäger, M.: Advanced MERIS Channel 11 Smile Correction as aBasis for Cloud Top Height Estimation Developed from SCIA-TRAN Radiative Transfer Calculations in the O2-A AbsorptionBand, Master’s thesis, University of Bremen, Bremen, Germany,2013.

Kalnay, E., Kanamitsu, M., Kistler, R., Collins, W., Deaven, D.,Gandin, L., Iredell, M., Saha, S., White, G., Woollen, J., Zhu, Y.,Leetmaa, A., Reynolds, B., Chelliah, M., Ebisuzaki, W., Higgins,W., Janowiak, J., Mo, K. C., Ropelewski, C., Wang, J., Jenne, R.,and Joseph, D.: The NCEP/NCAR 40-year reanalysis project, B.Am. Meteorol. Soc., 77, 437–470, https://doi.org/10.1175/1520-0477(1996)077<0437:TNYRP>2.0.CO;2, 1996.

Key, J. and Barry, R.G.: Cloud cover analysis with Arctic AVHRRdata. 1. Cloud detection, J. Geophys. Res., 94, 18521–18535,1989.

Kokhanovsky, A. A.: Cloud Optics, edited by: Mysak, L. A. andHamilton, K., Springer, Dordrecht, the Netherlands, 2006.

Kokhanovsky, A. A., von Hoyningen-Huene, W., and Burrows, J. P.:Determination of the cloud fraction in the SCIAMACHY groundscene using MERIS spectral measurements, Int. J. Remote Sens.,30, 6151–6167, https://doi.org/10.1080/01431160902842326,2009.

Krijger, J. M., Tol, P., Istomina, L. G., Schlundt, C., Schrijver, H.,and Aben, I.: Improved identification of clouds and ice/snowcovered surfaces in SCIAMACHY observations, Atmos. Meas.Tech., 4, 2213–2224, https://doi.org/10.5194/amt-4-2213-2011,2011.

Lindstrot, R., Preusker, R., and Fischer, J.: Empirical Cor-rection of Stray Light within the MERIS Oxygen A-Band Channel, J. Atmos. Ocean. Tech., 27, 1185–1194,https://doi.org/10.1175/2010JTECHA1430.1, 2010.

https://doi.org/10.5194/amt-13-6459-2020 Atmos. Meas. Tech., 13, 6459–6472, 2020

6472 L. Istomina et al.: Improved cloud detection over sea ice and snow during Arctic summer

Liu, Y., Key, J. R., Frey, R. A., Ackerman, S .A., and Menzel, W.P.: Nighttime polar cloud detection with MODIS, Remote Sens.Environ., 92, 181–194, 2004.

Lotz, W. A., Vountas, M., Dinter, T., and Burrows, J. P.: Cloudand surface classification using SCIAMACHY polarizationmeasurement devices, Atmos. Chem. Phys., 9, 1279–1288,https://doi.org/10.5194/acp-9-1279-2009, 2009.

Lyapustin, A. and Wang, Y.: The time series technique for aerosolretrievals over land from MODIS, Satellite Aerosol RemoteSensing over Land, edited by: Kokhanovsky A. A. and de Leeuw,G., Springer Praxis Publ., Chichester, UK, 69–99, 2009.

Lyapustin, A., Wang, Y., and Frey, R.: An automatedcloud mask algorithm based on time series of MODISmeasurements, J. Geophys. Res., 113, D16207,https://doi.org/10.1029/2007JD009641, 2008.

Madhavan, S., Angal, A., Dodd, J., Sun, J., and Xiong, X.: Ana-log and digital saturation in the MODIS reflective solar bands,Proc. SPIE 8510, Earth Observing Systems XVII, 85101N,https://doi.org/10.1117/12.929998, 2012.

Marks, H.: Investigation of Algorithms to Retrieve Melt Pond Frac-tion on Arctic Sea Ice from Optical Satellite Observations, Mas-ter’s thesis, Universität Tübingen, Tübingen, Germany, 2015.

Martins, J. V., Tanré, D., Remer, L., Kaufman, Y., Mattoo, S., andLevy, R.: MODIS Cloud screening for remote sensing of aerosolsover oceans using spatial variability, Geophys. Res. Lett., 29,8009, https://doi.org/10.1029/2001GL013252, 2002.

Minnis, P., Chakrapani, V., Doelling, D. R., Nguyen, L., Palikonda,R., Spangenberg, D. A., Uttal, T., Arduini, R. F., and Shupe,M.: Cloud coverage and height during FIRE ACE derived fromAVHRR data, J. Geophys. Res., 106, 15215–15232, 2001.

Nicolaus, M., Katlein, C., Maslanik, J., and Hendricks,S.: Changes in Arctic sea ice result in increasing lighttransmittance and absorption: LIGHT IN A CHANG-ING ARCTIC OCEAN, Geophys. Res. Lett., 39, L24501,https://doi.org/10.1029/2012GL053738, 2012.

Preusker, R. and Lindstrot, R.: Remote Sensing of Cloud-Top Pres-sure Using Moderately Resolved Measurements within the Oxy-gen A Band – A Sensitivity Study, J. Appl. Meteorol. Clim., 48,1562–1574, https://doi.org/10.1175/2009JAMC2074.1, 2009.

Schlundt, C., Kokhanovsky, A. A., von Hoyningen-Huene, W.,Dinter, T., Istomina, L., and Burrows, J. P.: Synergetic cloudfraction determination for SCIAMACHY using MERIS, At-mos. Meas. Tech., 4, 319–337, https://doi.org/10.5194/amt-4-319-2011, 2011.

Schröder, D., Feltham, D., Flocco, D., and Tsamados, M.:September Arctic sea-ice minimum predicted by springmelt-pond fraction, Nat. Clim. Change 4, 353–357,https://doi.org/10.1038/NCLIMATE2203, 2014.

Spangenberg, D. A., Chakrapani, V., Doelling, D. R., Minnis, P., andArduini, R. F.: Development of an automated Arctic cloud maskusing clear-sky satellite observations taken over the SHEBA andARM NSA sites, Proc. 6th Conf. on Polar Meteor., and Oceanog-raphy, 14–18 May 2001, San Diego, CA, USA, 246–249, 2001.

Trepte, Q., Arduini, R. F., Chen, Y., Sun-Mack, S., Minnis, P., Span-genberg, D. A., and Doelling, D. R.: Development of a daytimepolar cloud mask using theoretical models of near-infrared bidi-rectional reflectance for ARM and CERES, Proc. AMS 6th Conf.Polar Meteorology and Oceanography, San Diego, CA, USA, 4–18 May 2001, 242–245, 2001.

Warren, S. G.: Optical Properties of Snow, Rev. Geophys., 20, 67–89, 1982.

Wiebe, H., Heygster, G., Zege, E., Aoki, T., and Hori,M.: Snow grain size retrieval SGSP from optical satel-lite data: Validation with ground measurements and detec-tion of snow fall events, Remote Sens. Environ., 128, 11–20,https://doi.org/10.1016/j.rse.2012.09.007, 2013.

Zege, E., Malinka, A., Katsev, I., Prikhach, A., Heygster, G., Istom-ina, L., Birnbaum, G., and Schwarz, P.: Algorithm to retrieve themelt pond fraction and the spectral albedo of Arctic summer icefrom satellite optical data, Remote Sens. Environ., 163, 153–164,https://doi.org/10.1016/j.rse.2015.03.012, 2015.

Atmos. Meas. Tech., 13, 6459–6472, 2020 https://doi.org/10.5194/amt-13-6459-2020